Superbench 2.1 returning brief · original 1.x prompt and repository-run protocol. Existing artifacts are not relabeled as tool-free 2.1 generations. Review the suite →

Focused model compare

Low-Poly Tower Defense

GLM 5.3 is the starting run, compared with GPT-6.2 Astra, Claude Fable 5.2, GPT-6.2 Sol.